Screenpipe

let AI know what you are doing | backed by @ycombinator | npx screenpipe record | or download today 👇

Screenpipe is a local-first desktop application that continuously records what you see, hear, and say on your computer, then turns that captured activity into a searchable memory, written SOPs, and agents that can run repetitive work. It's aimed at knowledge workers and teams who want to stop losing context across tools, along with developers who want to feed real work history into AI assistants like Cursor or Claude Code. The launch matters now because Screenpipe joined YC S26 in May and recently released an alpha PII model that runs at 9ms on consumer devices, part of a push to keep the whole recording pipeline on-device , addressing the biggest objection to always-on screen capture tools. Under the hood, it listens for meaningful events like app switches, clicks, and typing pauses and captures a screenshot only when something changes, pairing each capture with the OS accessibility tree and falling back to OCR when that data isn't available . Audio from both system output and microphone is transcribed in real time using a local Whisper Large-V3-Turbo model or Deepgram in the cloud, with speaker diarization across Zoom, Meet, Teams, or any other app . Automations, which the team calls pipes, are triggered by patterns in that activity, and the tweet points to 20K+ GitHub stars, 1,900+ forks, and 130+ contributors as evidence of traction. The company is built by Mediar, Inc., founded in 2024 in San Francisco, with Louis Beaumont as sole founder . It positions itself as a source-available alternative to Rewind, Microsoft Recall, Granola, and Otter.ai that users can fully audit , which is the wedge for buyers who like the productivity story of always-on recall but can't send screen contents to a vendor cloud.

Tags

Comments (8)